logical indexing is usually faster than find

      j=find(si);
      s1=f(j);

logical indexing is usually faster than find,What does this mean,please give solution

[Jan]

Did you search for "logical indexing" in the documentation already? Even searching in the net for "Matlab logical indexing" will find many answers.

Answer 1

[Cedric]

There is a little overhead because of the function call and because FIND does a few operations that are unnecessary in most cases where you can use logical indexing.

When you do something like

 >> a = [1, 7, 5, 2] ;
 >> idx = find(a > 4)
 idx =
     2     3
 >> a(idx) = 0
 a =
     1     0     0     2

you call the function FIND with an argument that is a vector of logicals that you could directly use for indexing. FIND returns the position on non-false elements in its argument, that you can also use for linear or subscript indexing but with the overhead induced by the "computation" of positions. Using directly the logicals generated by the relational operation (in the present case) avoids the call to FIND, as shown by the last line executed in the code below

 >> a = [1, 7, 5, 2] ;
 >> a > 4
 ans =
     0     1     1     0
 >> class(ans)
 ans =
     logical
 >> a(a > 4) = 0
 a =
     1     0     0     2

You could actually easily write a function that roughly does what find does (just to illustrate the principle, assuming 1D arg):

 function idx = myFind( x )
     allIdx = 1:numel(x) ;
     idx = allIdx(x ~= 0) ;
 end

Now my experience is (and I call FIND "the infamous FIND" for this reason when I teach MATLAB), that I see people usually using FIND because they don't understand logical indexing, and it sets them up to implementing slow, non-vector, solutions, involving multiple nested FOR loops. For example, you will often see people building statements like

 [r,c] = find(A > 4) ;   % Get row and column indices.
 for ii 1 : numel(r)
    for jj = 1 : numel(c)
        A(r(ii), c(jj)) = 0 ;
    end
 end

with e.g. A = rand(1e4), instead of simply doing the following

 A(A > 4) = 0 ;

In such case, FIND isn't that slow, but the double loop is!

[Chuck]

The double for loops don't do the same thing as A(A>4) = 0; The loops set many elements to zero and the loop block is a good example of code not doing what is intended.

[Cedric]

Errata (thanks Chuck): people often do

 [r,c] = find(A > 4) ;   % Get row and column indices.
 for k = 1 : numel(r)
     A(r(k), c(k)) = 0 ;
 end

Answer 2

[ChristianW]

n = 1e6;
r = rand(n,1);
f = randn(n,1)+10;

si = r>0.5;

tic
j = find(si); 
s1 = f(j); % with FIND
toc

tic
s2 = f(si); % with logical indexing
toc

all(s1==s2)